Landslide prediction method and device, electronic equipment and readable storage medium

Abstract

The application provides a landslide risk early warning method, device, electronic equipment and readable storage medium. The landslide prediction method comprises: acquiring state data of a slope surface; inputting the state data into a landslide risk early warning model to obtain landslide risk data of the slope surface, wherein the landslide risk early warning model comprises a risk factor model and a sensor data model, the risk factor model is a model taking the state data of the slope surface as input and taking a landslide probability of the slope surface as output, the sensor data model is a model taking the state data of the slope surface as input and taking an abnormal result of the state data of the slope surface and a prediction result of the state data as output, and the landslide risk data comprises the landslide probability of the slope surface, the abnormal result of the state data and the prediction result of the state data; and generating a landslide prediction result according to the landslide risk data. Thus, the application embodiment can improve the accuracy of landslide risk prediction.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of Internet of Things (IoT) technology, and in particular to a landslide prediction method, apparatus, electronic device, and readable storage medium. Background Technology

[0002] Landslides are a highly dangerous natural disaster. Current methods for predicting landslides usually require investigation and analysis of the geological structure, and the results are used to determine the risk of landslides. However, this method has low accuracy in predicting landslide risk. Summary of the Invention

[0003] This invention provides a landslide prediction method, apparatus, electronic device, and readable storage medium to address the problem of low accuracy in landslide risk prediction.

[0004] To solve the above problems, the present invention is implemented as follows:

[0005] In a first aspect, embodiments of the present invention provide a landslide prediction method, comprising the following steps:

[0006] Obtain slope status data;

[0007] The state data is input into a landslide risk early warning model to obtain landslide risk data for the slope. The landslide risk early warning model includes a risk factor model and a sensor data model. The risk factor model is a model that takes the slope state data as input and outputs the probability of landslide occurrence. The sensor data model is a model that takes the slope state data as input and outputs abnormal results of the slope state data and prediction results of the state data. The landslide risk data includes the probability of landslide occurrence, abnormal results of the state data, and prediction results of the state data.

[0008] Landslide prediction results are generated based on the landslide risk data.

[0009] In some embodiments, before inputting the state data into the landslide risk early warning model to obtain the landslide risk data of the slope, the method further includes:

[0010] Obtain historical landslide data;

[0011] The historical landslide data is divided into training set data and test set data;

[0012] Using the slope state from the historical landslide data as input and the number of landslide occurrences as the regression target, a risk factor model is obtained by training the model using the training set data and the test set data.

[0013] In some embodiments, the sensor data model includes a time series data model for generating the prediction results, and before inputting the state data into the landslide risk early warning model to obtain the landslide risk data of the slope, it further includes:

[0014] Obtain historical displacement change data of the slope;

[0015] Predict the predicted displacement change at the target time point based on the first data in the historical displacement change data.

[0016] The actual displacement change at the target time point is obtained based on the historical displacement change data.

[0017] A time series data model is obtained by training the model based on the predicted displacement change and the actual displacement change.

[0018] The above steps are performed periodically and iteratively to update the time series data model.

[0019] In some embodiments, inputting the state data into a landslide risk early warning model to obtain landslide risk data for the slope includes:

[0020] Obtain historical state data of the slope;

[0021] Calculate the historical distribution parameters of the historical state data, wherein the historical distribution parameters include one or more of the mean and standard deviation of the historical state data;

[0022] The current landslide probability corresponding to the state data is obtained by inputting the state data into the risk factor model.

[0023] The parameter distribution range is determined based on the current landslide probability and the historical distribution parameters;

[0024] Abnormal data in the current state data is determined based on the distribution of the state data within the parameter distribution range.

[0025] In some embodiments, acquiring the slope state data includes:

[0026] The status data periodically transmitted by sensors installed on the slope surface is acquired, wherein the sensors include displacement sensors.

[0027] In some embodiments, the sensor further includes one or more of a groundwater sensor, a rainfall sensor, a tilt sensor, and a soil moisture sensor.

[0028] In some embodiments, the landslide risk early warning method is applied to an edge server, which is connected to the sensor via a narrowband Internet of Things (IoT) connection.

[0029] Secondly, embodiments of the present invention provide a landslide prediction device, comprising:

[0030] The acquisition module is used to acquire the state data of the slope.

[0031] An input module is used to input the state data into a landslide risk early warning model to obtain landslide risk data of the slope. The landslide risk early warning model includes a risk factor model and a sensor data model. The risk factor model is a model that takes the state data of the slope as input and outputs the probability of landslide occurrence. The sensor data model is a model that takes the state data of the slope as input and outputs abnormal results of the state data and prediction results of the state data. The landslide risk data includes the probability of landslide occurrence, abnormal results of the state data, and prediction results of the state data.

[0032] The generation module is used to generate landslide prediction results based on the landslide risk data.

[0033] Thirdly, embodiments of the present invention provide an electronic device, including: a transceiver, a memory, a processor, and a program stored in the memory and executable on the processor; the processor is configured to read the program in the memory to implement the steps in the landslide risk warning method as described in any one of the first aspects.

[0034] Fourthly, embodiments of the present invention provide a readable storage medium for storing a program, which, when executed by a processor, implements the steps of the landslide risk warning method as described in any one of the first aspects.

[0035] The landslide prediction method in this embodiment of the invention includes acquiring slope state data; inputting the state data into a landslide risk early warning model to obtain landslide risk data for the slope, wherein the landslide risk early warning model includes a risk factor model and a sensor data model. The risk factor model is a model that takes the slope state data as input and outputs the probability of a landslide occurring on the slope. The sensor data model is a model that takes the slope state data as input and outputs abnormal results of the slope state data and prediction results of the state data. The landslide risk data includes the probability of a landslide occurring on the slope, abnormal results of the state data, and prediction results of the state data. A landslide prediction result is generated based on the landslide risk data. Thus, this embodiment of the invention, by setting a risk factor model to predict the probability of a landslide and a sensor data model to predict data and analyze abnormal data, and by comprehensively combining the structures of each model to obtain landslide risk, can improve the accuracy of landslide risk prediction. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of the hardware structure provided in the embodiments of the present invention;

[0038] Figure 2 This is a schematic diagram of the sampling device in an embodiment of the present invention;

[0039] Figure 3 This is a flowchart of the landslide risk early warning method provided in the embodiments of the present invention;

[0040] Figure 4 This is a schematic diagram illustrating the training of the risk factor model in an embodiment of the present invention;

[0041] Figure 5 This is a data processing flowchart in an embodiment of the present invention;

[0042] Figure 6 This is a structural diagram of the landslide risk early warning model according to an embodiment of the present invention;

[0043] Figure 7 This is a schematic diagram of slope displacement in an embodiment of the present invention;

[0044] Figure 8 This is a schematic diagram illustrating the training of a time series data model in an embodiment of the present invention;

[0045] Figure 9 This is a schematic diagram of displacement change prediction in an embodiment of the present invention.

[0046] Figure 10 This is a schematic diagram of data distribution in an embodiment of the present invention;

[0047] Figure 11 This is a structural diagram of the landslide risk early warning device provided in an embodiment of the present invention;

[0048] Figure 12 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] In the embodiments of this invention, the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices. Additionally, the use of "and / or" in this application indicates at least one of the connected objects, such as A and / or B and / or C, representing seven possibilities: including A alone, B alone, C alone, and the presence of both A and B, both B and C, both A and C, and the presence of A, B, and C.

[0051] This invention provides a landslide prediction method.

[0052] like Figure 1 As shown, the hardware structure of the technical solution of the present invention mainly includes an edge server and sensors. The edge server is communicatively connected to the sensors. The sensors are directly installed at different locations on the slope to collect slope state data. The server analyzes the slope state data to obtain the landslide risk of the slope. This embodiment exemplarily shows two slopes. In practice, different areas can be classified into different slopes as needed, and their landslide risks can be analyzed separately.

[0053] In some embodiments, a cloud platform is also included, which is connected in communication with the edge servers. The cloud platform can aggregate the prediction structures of each edge server and send early warning information as needed.

[0054] In some embodiments, the sensor needs to include a displacement sensor, and the sensor may also include one or more of a groundwater sensor, a rainfall sensor, a tilt sensor, and a soil moisture sensor.

[0055] The sensor itself can be an existing or improved sensor, and the specific structure of the sensor is not further limited in this embodiment.

[0056] In some embodiments, the edge server communicates with sensors via narrowband Internet of Things (NB-IoT), which helps reduce power consumption. In some embodiments, each sensor periodically collects data, caches the collected data centrally, and then uploads it periodically to reduce network frequency, thereby reducing power consumption. This helps to achieve long-term data collection, reducing maintenance frequency and costs. The collected data is stored in the edge server for extended periods for later use.

[0057] Based on NB-IoT technology, it can achieve wide coverage, low power consumption, low cost, and large connectivity, making it very suitable for outdoor large-scale sensor deployment scenarios such as landslide early warning.

[0058] The sensors are deployed on-site to cover the entire monitored slope. The specific placement can be determined with the advice of professionals, and no further restrictions are made here.

[0059] Sensors can be integrated with a microcontroller (MCU). The MCU generates a timed interrupt to trigger sampling, which is then converted, packaged into report data, and wirelessly transmitted using an NB-IoT module. Multiple sensors can be integrated onto a single microcontroller.

[0060] The MCU with integrated sensors can communicate with the edge server via a wireless network. On one hand, it can upload collected data and its own status; on the other hand, it can receive and respond to control commands issued by the edge server. The MCU only receives commands from the edge server via the network while actively reporting data. After the transmission is completed, it enters sleep mode and does not maintain a long-term connection with the server.

[0061] The MCU selected supports low-power operation. Under most circumstances, the MCU sleeps in low-power mode. After a timed interrupt triggers data acquisition, the sampled data can be cached or uploaded according to the configuration. Power is supplied through a combination of battery and solar power, which helps improve the sensor's battery life.

[0062] The sampling and reporting frequency of the sensor can be set as needed. When the detection data is normal, the sampling and reporting frequency can be controlled to be relatively low, which helps to extend the standby time. When an anomaly occurs, such as when the edge server judges that the data is abnormal, the sampling and reporting frequency can be increased, which helps to improve the real-time performance of the data and the interaction between the end and the edge, and ensures timely response to the issued instructions.

[0063] like Figure 2 As shown in the figure below, the overall design of the sampling device mainly includes a microcontroller (MCU), a storage module, a power management module (powered by battery or solar power), a communication module (UART), and multiple sensors. The MCU samples the signals from each sensor at regular intervals according to the configured sampling frequency. After digital conversion and AD quantization, the sampled values, timestamps, and sampling device ID information are uploaded to the edge server via the NB-IoT module using transmission protocols such as CoAP (a transmission protocol).

[0064] While uploading sampling data each time, the MCU can report its own status, such as operating status, power consumption, network status, etc., to provide the edge server with information on the working status of the sampling device.

[0065] If the server fails to receive information uploaded by the device for a continuous period of time, it considers the device to be in an abnormal working state, which can trigger an alert and display the risk level as abnormal, so that staff can be notified to go to the site to troubleshoot the equipment.

[0066] When data upload is temporarily unavailable due to network fluctuations or other issues, the sampled data will be saved to the storage module and uploaded again once the network is restored. The sensor has a low sampling frequency and low power consumption on the device side, and is powered by batteries and solar energy.

[0067] During each data upload communication, the device can simultaneously receive information pushed by the edge server, forming a feedback loop of reporting, decision-making, and control. Under normal circumstances, the device can operate in low-power mode, with low-frequency sampling and data upload triggered by timer interrupts. When the edge server predicts an abnormal or imminent landslide risk, it can notify the device to exit low-power mode, increase the sampling frequency, and be ready to interact with the edge server at any time, improving the system's real-time response. In special circumstances, the edge server can proactively issue commands to drive sensors to sample and report current real-time indicators, supporting users in manual investigation and decision-making, and reducing false alarms caused by noise.

[0068] Edge servers, which can operate as servers or small clusters, manage slopes, personnel, and sensors, and are responsible for data sensing, analysis, and decision-making. Simultaneously, edge servers maintain communication with the cloud platform, exchanging model data and early warning information.

[0069] The edge server communicates with the sensors, managing sensor access, data transmission, and control. It acts as the server for sensor communication, where sampled data from the slope converges, and control commands sent to the sensors are also issued by the edge server. All sensors on the same slope belong to the same edge server. One edge server can simultaneously manage one or more slopes, maintaining the association information between the slopes and their corresponding sensors, forming a tree structure of gateway—slope—sensor.

[0070] The edge server manages slope-related information. It maintains information such as the slope's geographical location, geological features, sensor deployment status, and risk warnings. This information can be used for visualization and is also partially used in slope landslide risk assessment calculations. This information is entered by staff when adding new slopes for monitoring.

[0071] Edge servers can also be used for management personnel information. Each slope's detection corresponds to a specific staff member, maintaining the association between slopes and personnel. The access rights for slope sensor access / maintenance, data acquisition control, data display, and risk warnings are controlled by the relevant personnel.

[0072] Staff can interact with the edge server to query various information such as slope surface information, personnel management information, sensor association information, risk warning information, sensor data, and sensor operating status, gaining a comprehensive understanding of the slope condition.

[0073] Edge servers are also used to analyze, store, and visualize sensor sampling data, and to quantify landslide risk using models. The edge servers store data reported from the slope, periodically analyze it using models, and maintain the risk quantification results. They can also display sensor time-series data curves, risk quantification, and other information.

[0074] When the risk of landslides is high, the edge server notifies staff in an appropriate manner to respond promptly. The edge server can interact with the cloud platform to update risk quantification models and report early warnings of potential hazards.

[0075] A cloud platform is a SaaS (Software-as-a-Service) platform composed of server clusters. It provides unified management and support for edge servers. The cloud platform aggregates data from various edge servers, allowing each edge server to decide which information to synchronize to the cloud platform based on its specific needs. Due to the differences in sensor selection and configuration across different slopes, directly connecting and managing all sensors on the cloud platform is difficult and inefficient. Data access at the edge effectively solves the heterogeneity problem.

[0076] Meanwhile, the sensor data models for each slope can also be customized. Managing data and models at the edge allows for better matching to the actual scenario. By setting up edge servers, standardized components can be implemented on the cloud platform, while customization is achieved at the edge servers, connecting to on-site sensors and models. This enables efficient decision-making and real-time response, while also reducing the load on the cloud platform through distributed processing. Raw data can be further processed and anonymized on the edge servers before being aggregated to the cloud platform, ensuring data privacy and security.

[0077] During implementation, edge servers can report data to the cloud platform for storage and analysis, and can also use the computing resources of the cloud platform to update various models as needed.

[0078] The cloud platform manages edge servers, which can be seen as distributed extensions of the cloud platform. Risk alerts can be viewed as applications running on edge servers, and the cloud platform manages the entire application lifecycle. The cloud platform also enables information fusion, allowing for a higher-level understanding of the overall situation. Utilizing data from various edge servers, the cloud platform conducts further statistical analysis at a higher level, iteratively updating and pushing necessary models to the edge servers, thus facilitating model optimization.

[0079] This proposal adopts an edge platform / cloud platform deployment approach, which is more flexible, reliable, feasible, and cost-effective. It not only alleviates traffic pressure but also offers significant advantages in data security, data privacy, and personalization.

[0080] like Figure 3 As shown, in some embodiments, the landslide prediction method includes the following steps:

[0081] Step 301: Obtain the slope state data.

[0082] In this embodiment, the slope state data refers to geological parameters such as lithology, hydrology, climate, soil quality, vegetation, and slope of various slopes collected using the aforementioned sensors.

[0083] Status data can be collected periodically and transmitted in real time. In some embodiments, step 101 may also include: acquiring status data periodically transmitted by sensors positioned on the slope. Periodically transmitting status data helps improve the battery life of the sampling device.

[0084] Step 302: Input the status data into the landslide risk early warning model to obtain the landslide risk data of the slope.

[0085] In this embodiment, the landslide risk early warning model includes a risk factor model and a sensor data model.

[0086] The risk factor model is a model that takes the state data of the slope as input and the probability of landslides as output. By inputting the obtained state data into the risk factor model, the predicted probability of landslides can be obtained.

[0087] like Figure 4 As shown, this risk factor model is obtained through pre-trained model training.

[0088] In some embodiments, the steps of training a model to obtain a risk factor model include:

[0089] Obtain historical landslide data;

[0090] The historical landslide data is divided into training set data and test set data;

[0091] Using the slope state from the historical landslide data as input and the number of landslide occurrences as the regression target, a risk factor model is obtained by training the model using the training set data and the test set data.

[0092] In this embodiment, historical landslide data includes previously detected slope condition data and the results of landslides.

[0093] In this embodiment, training data is generated based on the historical landslide data. For example... Figure 5 As shown, in this embodiment, the historical landslide data is first preprocessed, and then the preprocessed data is used for model training. During the preprocessing process, the original data is first screened. It should be understood that the selected historical landslide data mainly consists of state data corresponding to the time period in which the landslide occurred. Obviously, model training also requires the selection of some data under normal conditions, which are referred to as normal data in this embodiment.

[0094] The next step is to clean the data. After selecting the required historical landslide data, the data can be further improved. Specifically, this can include deduplicating duplicate data and repairing missing or incorrectly formatted data.

[0095] The next step is to aggregate the processed data. For example, the data can be aggregated by time (exposure period) and location (distinguishing key) to obtain landslide data from various locations across different time periods.

[0096] Next, the data in the set is further processed. Specifically, each feature is used as a factor and encoded separately. For example, one-bit effective One-Hot encoding can be used. Discrete data can be processed directly, while continuous data (such as slope) can be discretized first and then processed. The regression target of the statistical data is the number of landslides.

[0097] After processing the landslide data, a dataset is created and divided into training and testing sets. The training set is used directly for model training, while the testing set is used to test the trained model. The amounts of training and testing data can be set as needed. Generally, the training set has more data than the testing set. In this embodiment, the ratio of training to testing data is set to 7:3. Obviously, the specific amounts can be set as needed in practice, and this embodiment does not further limit or describe them. This yields training data including both training and testing sets.

[0098] Finally, the parameters of the generalized linear model (GLM) are estimated and evaluated.

[0099] In this embodiment, landslide frequency is roughly assumed to follow a Poisson distribution. Slope geological characteristics are used as risk factors, and the regression coefficients of factors are calculated using the Geometric Matrix (GLM), followed by testing for evaluation. The GLM performs maximum likelihood regression of each factor coefficient based on the distribution assumption, typically using iterative numerical calculation methods. In practice, for each feature dimension, the most common term is selected as the baseline (intercept), and that column is removed from the feature matrix. The corresponding factor weights are then set to 1. The feature matrix formed from the training set data and the target vector are used to fit and regress the GLM coefficients. The regression result includes the baseline cutoff term and the coefficients corresponding to each feature dimension. For prediction, the intercept value is multiplied by the coefficients corresponding to each feature to obtain the result.

[0100] The evaluation of a model can be quantified using the mean squared error (MSE) between predicted and actual values ​​in the test set data; the smaller the MSE, the more accurate the model. During prediction, the GLM provides confidence interval information. The confidence interval is directly related to the data used for modeling; generally, the larger the amount of effective data, the more accurate the model and the narrower the confidence interval.

[0101] The GLM model, built using a large number of samples, can quantify the prior risk of a slope. By simply inputting the features corresponding to the slope into the model, the frequency of landslides can be predicted. The higher the frequency, the higher the actual risk of a landslide.

[0102] The model training process can be understood as taking the slope state in the historical landslide data as input, the number of landslides as the regression target, and obtaining the risk factor model through the training set data and the test set data.

[0103] During training, test data is first input into the training model. For example, if historical landslide data that has occurred is input into the model, the expected output value should be "landslide occurred". If normal data is input into the model, the expected output value should be "no landslide occurred". The training data contains the true values ​​corresponding to the input data. By comparing the true values ​​with the expected values ​​and constructing a loss function, the model parameters are adjusted according to the loss value. After each adjustment of the model parameters, the model can be tested using test set data and then further trained. In this way, the model is continuously iterated and optimized. Finally, the trained model is used as a risk factor model and deployed to various edge servers.

[0104] After the edge server receives the status data collected by each sensor, it inputs the status data into the model to generate a prediction result for landslide risk. The prediction result can then be further visualized or an early warning message can be issued as needed.

[0105] like Figure 4As shown, it should be understood that the above model training process can be completed in advance through offline tasks, and then the trained risk factor model can be deployed to the edge server. During use, the collected state data can be continuously used to refer to the above training process to continuously train and optimize the risk factor model, which helps to improve the accuracy of the prediction results of the risk factor model.

[0106] During the prediction process, the obtained state data can also be preprocessed using the above process, which helps to improve the accuracy of the prediction results of the risk factor model.

[0107] The sensor data model takes slope state data as input and outputs abnormal results of the slope state data and prediction results of the state data. The slippage risk data includes the probability of slope landslide, abnormal results of state data, and prediction results of slippage data.

[0108] like Figure 6 As shown, in some embodiments, the sensor data model can be understood as consisting of two models: specifically, the sensor data model consists of a time series data model and a statistical anomaly detection model.

[0109] The time-series data model takes slope condition data as input and outputs predictions based on that data. It predicts the trend of numerical changes over a future period based on sensor-collected condition data. When a preset threshold is reached, a high landslide risk is considered present. In essence, this time-series data model is used to predict the timing of potential landslides. This model primarily targets displacement monitoring values ​​because displacement generally exhibits a unidirectional trend over time. In special cases, such as localized variations in measurements due to temperature within a day, the overall trend remains one of gradually increasing displacement. By establishing the relationship between displacement and time, landslide risk can be predicted more directly.

[0110] In this embodiment, displacement can be measured using constant displacement sensors such as a wire displacement meter and an inclinometer. The wire displacement meter is mainly used to measure the surface displacement of a slope, while the inclinometer is often used to measure deep displacement.

[0111] like Figure 7 As shown, a wire displacement gauge is used as an example for illustration. Over time, the slope surface gradually moves under the influence of gravity. In the entire landslide development process, the cumulative displacement over time can generally be divided into three stages: initial deformation, constant-rate deformation, and accelerated deformation.

[0112] During the initial and constant-rate deformation stages, the slope is relatively stable, and the displacement changes tend to be gradual. In the accelerated deformation stage, the displacement changes suddenly at a faster rate until a landslide occurs. The original displacement signal is sampled approximately every few hours, and the sampling results are reported to the edge gateway via the network. The server stores the raw data and periodically initiates modeling tasks. During modeling, historical data related to the sensors is extracted; the data is in the form of time-series data where x represents the sampling time and y represents the displacement sampling value. The model is updated using data-driven parameter estimation. The updated model is evaluated and can be used for inference. The model outputs sampled predicted values ​​for a specified future time period, along with the corresponding confidence intervals.

[0113] In this embodiment, data can be continuously accumulated for model iteration, which helps improve prediction accuracy. This avoids overfitting problems caused by discrepancies between the modeled data and actual data distribution due to initial data scarcity. Iteration effectively improves the prediction accuracy of the sensor data model.

[0114] like Figure 8 As shown in the figure below, the overall process can be divided into two tasks: prediction and training.

[0115] In some embodiments, the training task mainly includes the following steps:

[0116] Obtain historical displacement change data of the slope;

[0117] Predict the predicted displacement change at the target time point based on the first data in the historical displacement change data.

[0118] The actual displacement change at the target time point is obtained based on the historical displacement change data.

[0119] A time series data model is obtained by training the model based on the predicted displacement change and the actual displacement change.

[0120] The above steps are performed periodically and iteratively to update the time series data model.

[0121] In this embodiment, a certain amount of historical displacement change data can be obtained first, and the displacement change data can be used to train the model to obtain a time series data model.

[0122] The edge gateway periodically starts the training task, queries the data required for modeling, establishes training and testing datasets, and begins model fitting using the training set. The model training process can be summarized as follows: taking displacement as input and the predicted displacement results for one or more subsequent target time points as output, the model is trained. The model's differences are adjusted based on the discrepancy between the predicted values ​​and the actual values ​​in the historical displacement change data (i.e., the difference between the predicted displacement change and the actual displacement change). A loss function is constructed, the model is trained, and the trained model is used as the time series data model.

[0123] like Figure 8 As shown, during use, sampling values ​​are continuously obtained using displacement sensors. While the sampling values ​​are uploaded, the raw data is also stored locally on the edge server. The stored raw data is used as new historical displacement change data to retrain the time series data model. In this way, the model is continuously evaluated and updated, and the displacement prediction results are obtained through the time series data model. The prediction results can be compared with the subsequently collected real values, and the model can be further trained again based on the comparison results. This iterative process allows for continuous optimization, updating, and improvement of the time series data model during use.

[0124] It should be understood that the above training tasks can be performed on edge servers. If computing resources are limited, the edge server can also submit the training task to the cloud platform for computation, and then send it back to the edge server after the computation is completed.

[0125] In some embodiments, the time series data model may employ an exponential decay model to fit the cumulative displacement, and the time series data model can be summarized by the following formula:

[0126]

[0127] In the above formula, h(t) is the cumulative displacement, φ is the intercept term, θ and β are random parameters that determine the evil slope of the model, θ is a log-normal distribution, and β is a Gaussian distribution. At each time step, θ and β will be updated based on the most recent sampled observation posterior of h(t), and ∈ is Gaussian white noise, satisfying ∈ ~ N(0, σ 2 ), satisfy:

[0128] E[h(t)|θ,β]=φ+θexp(βt).

[0129] Time series data models are fitted using historical data, and predictions are made with each new data entry, while the model's posterior is updated. As data accumulates and iterates, the model's predictions become increasingly accurate. Each displacement sensor can be pre-set with a warning threshold; the time series data model can then estimate how long it will take for the current distance to reach that threshold. Statistical analysis of the estimated times for each sensor model yields an estimate of the landslide's duration.

[0130] In practice, a weighted average can be used to statistically analyze the estimated time corresponding to each sensor model. The weighting is related to the accuracy of the model evaluation. The more accurate the model, the higher its relative weight, which helps to improve the estimation accuracy of landslide time.

[0131] When utilizing displacement-time curves and their prediction sequences, two main aspects are considered: first, whether the predicted value over a future period reaches a threshold level; if the predicted value exceeds a pre-set threshold, the risk level is considered abnormal; second, the displacement-time curve of the current accumulated data, when it is found that the end enters an accelerated deformation stage, is considered to have an abnormal risk level or be on the verge of slippage. In implementation, the data can be resampled to unify the time step, and the curve can be normalized first using the uniform deformation stage within the curve.

[0132] During the uniform deformation stage, the displacement velocity is v, and the displacement S = v * t. By dividing the displacement S by the velocity v, the vertical axis of the St curve can be transformed to the same time dimension as the horizontal axis.

[0133]

[0134] Please also refer to Figure 9 , where ΔS(i) is the change in slope displacement within a certain unit time period, which can be a day, a week, or other different time periods; v is the displacement rate during the constant velocity deformation stage; T(i) is the ordinate with the same dimension as time after transformation.

[0135] Please continue reading. Figure 9 , Where, α i ΔT is the slope of the tangent at the end of the curve, ΔT is the change in the vertical coordinate, and Δt is the change in the horizontal coordinate. In this embodiment, when entering the accelerated deformation stage, the tangent slope is considered to be abnormal and the risk level is considered to be near slippage when the tangent slope exceeds 75 degrees.

[0136] The output of time series data models can be further visualized to serve as early warning data for disaster response.

[0137] The statistical anomaly detection model takes slope condition data as input and outputs anomalies in that data. It primarily detects anomalies in monitored values ​​of data such as soil moisture content, rainfall, and groundwater pore water pressure. Generally, these monitored values ​​fluctuate within normal ranges, are highly correlated with environmental changes, and are relatively random in time. When the detected values ​​fluctuate beyond normal levels, a higher landslide risk is considered.

[0138] In some embodiments, inputting the state data into a landslide risk early warning model to obtain landslide risk data for the slope includes:

[0139] Obtain historical state data of the slope;

[0140] Calculate the historical distribution parameters of the historical state data, wherein the historical distribution parameters include one or more of the mean and standard deviation of the historical state data;

[0141] The current landslide probability corresponding to the state data is obtained by inputting the state data into the risk factor model.

[0142] The parameter distribution range is determined based on the current landslide probability and the historical distribution parameters;

[0143] Abnormal data in the current state data is determined based on the distribution of the state data within the parameter distribution range.

[0144] For data such as temperature, pore water pressure, rainfall, and soil moisture content, the sampled values ​​vary irregularly over time, mainly influenced by random environmental factors such as meteorological processes. This type of data is generally affected by multiple factors and fluctuates within a normal range, such as... Figure 10 As shown in this embodiment, it is assumed that it roughly follows a normal distribution, and therefore an anomaly detection model is designed using its statistical indicators.

[0145] In this dormitory, the sampled values ​​of each sensor are statistically analyzed according to the time dimension to obtain the mean and standard deviation. For example, the sampled values ​​of each month in different years can be statistically analyzed by month. In this way, the climate of the same month is similar and has more reference value.

[0146] When conducting anomaly detection, the risk level of the slope is first determined using a risk factor model, and a threshold parameter k corresponding to that risk level is selected. Generally, the higher the landslide frequency predicted by the risk factors, the more sensitive the anomaly detection requirements, and the smaller the value of k.

[0147] Next, select the mean S1 and standard deviation S2 of the statistical indicators corresponding to the current time in the above time dimension, and compare whether the current sampled value S0 is within the range of [S1-kS2, S1+kS2]. If it exceeds the fluctuation range, issue a risk warning and the risk level is abnormal.

[0148] Step 303: Generate landslide prediction results based on the landslide risk data.

[0149] Based on the landslide risk data obtained above, a prediction of the current slope landslide risk can be obtained. For example, landslide risk can be classified into several risk levels such as normal, abnormal, and imminent landslide, according to different combinations of these risk levels. When the risk reaches a relatively high level such as abnormal or imminent landslide, early warning information can be further pushed to relevant personnel to respond promptly to potential dangers and ensure the safety of life and property.

[0150] This invention utilizes historical statistical data to estimate risk factors for multiple geological features of a slope, regresses to obtain the risk frequency of landslides, and combines this with data from field-deployed sensors. It also integrates statistical anomaly detection and time series prediction techniques to assess landslide risk. This combination improves prediction performance, and the generated early warning information includes estimated landslide timing information, further enhancing the accuracy of risk prediction.

[0151] This invention also provides a landslide prediction device.

[0152] like Figure 11 As shown, in one embodiment, the landslide prediction device 1100 includes:

[0153] Module 1101 is used to acquire slope status data;

[0154] Input module 1102 is used to input the state data into a landslide risk early warning model to obtain landslide risk data of the slope. The landslide risk early warning model includes a risk factor model and a sensor data model. The risk factor model is a model that takes the state data of the slope as input and outputs the probability of landslide occurrence. The sensor data model is a model that takes the state data of the slope as input and outputs abnormal results of the state data and prediction results of the state data. The landslide risk data includes the probability of landslide occurrence, abnormal results of the state data, and prediction results of the state data.

[0155] The generation module 1103 is used to generate landslide prediction results based on the landslide risk data.

[0156] In some embodiments, it also includes:

[0157] The historical landslide data acquisition module is used to acquire historical landslide data;

[0158] The dataset partitioning module is used to divide the historical landslide data into training set data and test set data;

[0159] The first training module is used to obtain a risk factor model by taking the slope state in the historical landslide data as input, taking the number of landslides as the regression target, and training the model using the training set data and the test set data.

[0160] In some embodiments, the sensor data model includes a time series data model for generating the prediction results, and further includes:

[0161] The historical displacement change acquisition module is used to acquire historical displacement change data of the slope.

[0162] The prediction module is used to predict the predicted displacement change at a target time point based on the first data in the historical displacement change data.

[0163] The actual displacement change acquisition module is used to acquire the actual displacement change at the target time point based on the historical displacement change data.

[0164] The second training module is used to train the model based on the predicted displacement change and the actual displacement change to obtain a time series data model.

[0165] An iterative module is used to periodically execute the above steps to update the time series data model.

[0166] In some embodiments, the input module 1102 includes:

[0167] The acquisition submodule is used to acquire historical state data of the slope.

[0168] The calculation submodule is used to calculate the historical distribution parameters of the historical state data, wherein the historical distribution parameters include one or more of the mean and standard deviation of the historical state data;

[0169] The input submodule is used to input the state data into the risk factor model to obtain the current landslide probability corresponding to the state data;

[0170] The interval determination submodule is used to determine the parameter distribution interval based on the current landslide probability and the historical distribution parameters;

[0171] The anomaly determination submodule is used to determine the abnormal data in the current state data based on the distribution status of the state data in the parameter distribution interval.

[0172] In some embodiments, the acquisition module 1101 is specifically used to acquire status data periodically transmitted by a sensor disposed on the slope, wherein the sensor includes a displacement sensor.

[0173] In some embodiments, the sensor further includes one or more of a groundwater sensor, a rainfall sensor, a tilt sensor, and a soil moisture sensor.

[0174] In some embodiments, the landslide risk early warning method is applied to an edge server, which is connected to the sensor via a narrowband Internet of Things (IoT) connection.

[0175] The landslide risk early warning device 1100 of this embodiment can implement all the steps of the above-described landslide risk early warning method embodiment and achieve basically the same technical effect, which will not be described again here.

[0176] This invention also provides an electronic device. Please refer to [link to relevant documentation]. Figure 12 The electronic device may include a processor 1201, a memory 1202, and a program 12021 stored in the memory 1202 and executable on the processor 1201.

[0177] When the electronic device is a terminal, program 12021 can be executed by processor 1201 to achieve the following: Figure 1 Any steps in the corresponding method embodiments and the achievement of the same beneficial effects will not be repeated here.

[0178] When the electronic device is a network-side device, program 12021 can be executed by processor 1201 to achieve the following: Figure 12 Any steps in the corresponding method embodiments and the achievement of the same beneficial effects will not be repeated here.

[0179] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by hardware related to program instructions, and the program can be stored in a readable medium.

[0180] This invention also provides a readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described functions. Figure 1 Any step in the corresponding method embodiment can achieve the same technical effect, and will not be repeated here to avoid repetition.

[0181] The storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0182] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A landslide prediction method, characterized in that, Includes the following steps: Obtain slope status data; The state data is input into a landslide risk early warning model to obtain landslide risk data for the slope. The landslide risk early warning model includes a risk factor model and a sensor data model. The risk factor model takes the slope state data as input and outputs the probability of a landslide. The sensor data model takes the slope state data as input and outputs abnormal results of the slope state data and prediction results of the state data. The landslide risk data includes the probability of a landslide, abnormal results of the state data, and prediction results of the state data. The risk factor model is obtained through pre-trained model training. A landslide prediction result is generated based on the landslide risk data; The sensor data model consists of a time series data model and a statistical anomaly detection model. The time series data model employs an exponential decay model to fit the cumulative displacement, and the time series data model is summarized by the following formula: ； In the formula, This is the cumulative amount of displacement. The intercept term for the constant term. and To determine the random parameters of the model slope, It follows a lognormal distribution. For a Gaussian-distributed distribution, at each time step, and According to The most recent post-abnormal update of the sampling observation. It is Gaussian white noise, satisfying , satisfy: ； The statistical anomaly detection model takes the slope state data as input and the anomaly results of the slope state data as output.

2. The landslide prediction method according to claim 1, characterized in that, Before inputting the state data into the landslide risk early warning model to obtain the landslide risk data of the slope, the process also includes: Obtain historical landslide data; The historical landslide data is divided into training set data and test set data; Using the slope state from the historical landslide data as input and the number of landslide occurrences as the regression target, a risk factor model is obtained by training the model using the training set data and the test set data.

3. The landslide prediction method according to claim 1, characterized in that, The sensor data model includes a time series data model for generating the prediction results. Before inputting the state data into the landslide risk early warning model to obtain the landslide risk data of the slope, it also includes: Obtain historical displacement change data of the slope; Predict the predicted displacement change at the target time point based on the first data in the historical displacement change data. The actual displacement change at the target time point is obtained based on the historical displacement change data. A time series data model is obtained by training the model based on the predicted displacement change and the actual displacement change. The above steps are performed periodically and iteratively to update the time series data model.

4. The landslide prediction method according to claim 1, characterized in that, The step of inputting the state data into the landslide risk early warning model to obtain the landslide risk data of the slope includes: Obtain historical state data of the slope; Calculate the historical distribution parameters of the historical state data, wherein the historical distribution parameters include one or more of the mean and standard deviation of the historical state data; The current landslide probability corresponding to the state data is obtained by inputting the state data into the risk factor model. The parameter distribution range is determined based on the current landslide probability and the historical distribution parameters; Abnormal data in the current state data is determined based on the distribution of the state data within the parameter distribution range.

5. The landslide prediction method according to any one of claims 1 to 4, characterized in that, The acquisition of slope state data includes: The status data periodically transmitted by sensors installed on the slope surface is acquired, wherein the sensors include displacement sensors.

6. The landslide prediction method according to claim 5, characterized in that, The sensor also includes one or more of the following: groundwater sensor, rainfall sensor, tilt sensor, and soil moisture sensor.

7. The landslide prediction method according to claim 6, characterized in that, The landslide prediction method is applied to an edge server, which communicates with the sensor via a narrowband Internet of Things (IoT) connection.

8. A landslide prediction device, characterized in that, include: The acquisition module is used to acquire the state data of the slope. An input module is used to input the state data into a landslide risk early warning model to obtain landslide risk data for the slope. The landslide risk early warning model includes a risk factor model and a sensor data model. The risk factor model takes the slope state data as input and outputs the probability of a landslide. The sensor data model takes the slope state data as input and outputs abnormal results of the slope state data and prediction results of the state data. The landslide risk data includes the probability of a landslide, abnormal results of the state data, and prediction results of the state data. The risk factor model is obtained through pre-trained model training. The generation module is used to generate landslide prediction results based on the landslide risk data; The sensor data model consists of a time series data model and a statistical anomaly detection model. The time series data model employs an exponential decay model to fit the cumulative displacement, and the time series data model is summarized by the following formula: ； In the formula, This is the cumulative amount of displacement. The intercept term for the constant term. and To determine the random parameters of the model slope, It follows a lognormal distribution. For a Gaussian-distributed distribution, at each time step, and According to The most recent post-abnormal update of the sampling observation. It is Gaussian white noise, satisfying , satisfy: ； The statistical anomaly detection model takes the slope state data as input and the anomaly results of the slope state data as output.

9. An electronic device, comprising: A transceiver, a memory, a processor, and a program stored in the memory and executable on the processor; characterized in that the processor is configured to read the program in the memory to implement the steps in the landslide prediction method as described in any one of claims 1 to 7.

10. A readable storage medium for storing a program, characterized in that, When the program is executed by the processor, it implements the steps in the landslide prediction method as described in any one of claims 1 to 7.

Images